Towards Effective MLLM Jailbr | Pangram Labs